Automate your work with AI
Every company has work that is the same every week, like retyping an invoice or answering the same question. We build the piece that takes that work over, and show you where it stops.
What this involves
Where it starts
We first look at which work keeps coming back and how much time it takes. Usually that is processing invoices, or email that always gets the same answer.
What a model does and does not do
A language model reads and writes text, and it guesses when it does not know something. So we never let it work on its own: the suggestion goes past you, or past a rule that checks the outcome.
The plumbing around it
The model is the smallest part. The work is in the link to your accounting, your mailbox or your stock, and in keeping track of what has already been handled.
Where it stops
Whatever the model is unsure about stays put, with the reason next to it. You then check one row instead of an outcome you can no longer do anything about.
How the work flows
An automation is not a button that takes everything over. It picks up one step, makes a suggestion about it, and hands it back to you when it is not sure.
-
What comes in
an invoice by email, a request through the form, or a new row in your system.
-
The model
reads it, pulls out the data and states how confident it is.
-
The check
a rule or a person reviews everything the model is unsure about.
-
Your system
receives the processed row, with the source underneath so you can trace it back.
What the model is unsure about does not go through; it goes on a list. That keeps one place to look instead of an outcome you have to verify everywhere.
For example
- Reading invoices and posting them to the right ledger account
- Pre-sorting email and drafting a reply
- Building quotes from data you already have
- Summarising documents and extracting the data
Where is the best place to start?
With work that comes back often and whose outcome you can check. Reading invoices is a good start: there are many of them, and you see straight away whether it is right.
Does our data go to an AI company?
That depends on which model we use. We say up front which model runs where and what data it sees, and we can pick a model that stays inside Europe.
What if the model gets it wrong?
That is what we design for. Anything the model is unsure about does not go through but lands on a list, with the source underneath so you can see where it went wrong.
Does this work with our accounting package?
If the package offers an integration or has a database we can reach, it works. In the intro call we work out which route fits.
Can we start small?
Yes. We automate one step and see what it gives you, and then you decide whether a second one follows.
Does this replace someone?
In practice the work shifts: the retyping goes and the checking stays. What that means for your company depends on how much of the work was retyping.
We also build
Which work comes back every week?
Name the step that costs you the most time, and we will say whether it can be automated and what it costs.